Google, IBM & Meta Certificates – 40% Off
One plan covers every Professional Certificate on Coursera.
Unlock All Certificates
Learn about Mowgli, a novel data-driven rate control system for real-time video conferencing that addresses the practical deployment challenges of machine learning-based approaches in this 16-minute conference presentation from NSDI '25. Discover how researchers from Princeton University and University of Illinois Urbana-Champaign developed an innovative solution that uses existing telemetry logs for training instead of requiring disruptive online learning that degrades user experience. Explore the technical approach that combines robust learning techniques including conservative reasoning about alternate behavior to minimize risk and richer model formulations to account for environmental noise, enabling the system to extract performant decisions from historical data despite timing and ordering issues. Examine the comprehensive evaluation results showing Mowgli's superior performance compared to the widely deployed GCC algorithm, achieving 15-39% increases in average video bitrates while reducing freeze rates by 60-100% across diverse network conditions including both emulated and real-world scenarios. Understand the broader implications for making data-driven rate control algorithms practically viable for production video conferencing services by eliminating the training-time performance penalties that have historically prevented their adoption.
